The objective of this paper is to develop an Artificial Neural Network (ANN) model to estimate soil temperature for any day. We used air average temperature, sunshine, radiation and soil temperature for meteorological data between years [1980 and 1984] at Nineveh/Iraq Meteorological Station.
In this research, three ANN models with their associated training algorithms (Backpropagation neural network (BPNN), Cascade-Forward and Nonlinear Autoregressive (NARX)) were used for estimating soil temperatures at different depths of 5, 10, 20, 50 and 100cm within the time 9, 12 and 15 respectively.
The performance of the three models with their training algorithms were compared with the measured soil temperature values to identify the best fit ANN model for soil temperature forecasting. The results showed that the NARX model is the best model. Finally, a comparison between five optimization ANN training algorithms was adopted to train NARX ANN model to identify best fit optimization algorithm for forecasting soil temperature with best results. From comparisons, TrainLM is the best optimization algorithm for training NARX model.